Table of contents

Storage & Data

A pod’s own disk is ephemeral — anything written inside the container that isn’t on an attached volume can be lost when the pod stops, is deleted, or is recreated. To keep datasets, model checkpoints, and outputs, put them on persistent storage.

Every Podstack image expects your persistent storage at /data. Mount a volume there and your work survives restarts and recreations.

# Typical layout inside a pod
/data/
├── datasets/      # input data
├── models/        # checkpoints, weights
└── output/        # results

Storage options

OptionBest forHow a pod uses it
NFS VolumeShared, persistent filesystem for podsMounted into the pod at a path (default /data)
Data VolumeStandalone disk (blank, or imported from a URL/registry/PVC)Attached as a persistent disk
Object StorageLarge datasets, artifacts, S3-compatible accessAccessed over the S3 API / CLI from inside the pod
File StorageOn-cluster S3 buckets with lower egressAccessed over the S3 API (may be gated per account)

For most QuickPod workflows you’ll attach an NFS Volume at /data. See the full Storage section for object and file storage.

Create an NFS Volume

You can create a volume ahead of time from the NFS Volumes page, or inline while launching a pod.

  1. Go to NFS Volumes and click Create Volume. (You need a selected project and a verified account.)
  2. In the Create NFS Volume dialog, fill in:
    • Volume Name — letters, numbers, and hyphens only (no dots or spaces).
    • Project — pre-selected (read-only).
    • Quota (GB) — choose No Limit or a size (10, 25, 50, 100, 250, 500, or 1000 GB). Default is 50 GB.
    • Billing Period — Monthly, Quarterly, or Annual.
    • Enable NFS access — leave checked so pods can mount it.
  3. Review the Estimated Cost panel (rate per GB, GST, and any period discount). The volume is billed upfront for the period.
  4. Click Create Volume.

A volume’s status is one of active, suspended, pending_deletion, deleted, or error. You can mount an active volume into any pod in the same project.

Billing is on the provisioned quota, not on how much you’ve written. The whole billing cycle is charged with no pro-rated refund if you delete early.

Attach a volume to a pod

During launch, in the Configure step’s NFS Volume card (see Launch a Pod):

  1. Select your volume from the list (or No Volume for ephemeral storage only).
  2. Set the Mount Path Inside Pod — keep the default /data unless you have a reason to change it.
  3. Launch the pod. The volume is mounted automatically when the pod starts.

Multiple pods in the same project can mount the same NFS volume at once — useful for sharing a dataset across parallel jobs.

ComfyUI note. For the ComfyUI image, don’t change the /data mount path — its custom nodes, models, and workflows live under /data and need to stay there to survive a restart.

Data Volumes

For a standalone disk (rather than a shared NFS export), use a Data Volume:

  1. Go to Data Volumes and click Create Volume.
  2. Step through the wizard: Project → Basic Info → Storage → Source → Review.
    • Basic Info — Volume Name (lowercase; letters, numbers, hyphens, dots) and an optional Description.
    • Storage — Size (GB) (minimum 1 GB), an optional Storage Class, and an Access Mode (ReadWriteOnce, ReadWriteMany, or ReadOnlyMany).
    • Source — Blank Disk, import from an HTTP/HTTPS URL, pull from a Container Registry, or clone an existing PVC.
  3. Finish the wizard to provision the volume.

Moving files in and out

Over SSH / SCP

If your pod has an SSH key attached, copy files with scp using the pod’s SSH endpoint:

scp -i ~/.ssh/<key_name> ./dataset.zip podstack@ssh-<subdomain>.cloud.podstack.ai:/data/datasets/

From object storage

Inside the pod, pull large datasets from an S3 bucket:

aws s3 cp s3://my-bucket/dataset.tar.gz /data/datasets/ --endpoint-url https://s3.podstack.ai

See Object Storage for buckets, access keys, and presigned URLs.

Browse outputs in your browser

To download generated files (renders, images, processed video) from /data/output/, expose a port and serve the directory with Python’s built-in HTTP server. See Accessing Output Files in the Browser.

What persists, and what doesn’t

  • Persists: anything on an attached NFS Volume or Data Volume, and anything in Object/File Storage. These survive pod stop, delete, and recreation.
  • May be lost: anything written to the container’s own disk outside a mounted volume — especially when you stop or delete the pod.

Rule of thumb: write temporary scratch files to local disk for speed, but write anything you want to keep to /data.

Next steps

  • Launch a Pod — attach a volume during launch.
  • MLOps — track experiments and register the models you produce.
  • Storage — object storage, file storage, and NFS in depth.